You know those Monday mornings when you sit down with your coffee, open your laptop, and realise you’re already behind? I had one of those recently during peak ski season. 47 unread emails, a dozen active Slack threads, 26 open tasks in Asana, and two project launches scheduled for the same week. I lead a remote team spread across several European countries, and we move more than 350,000 passengers through the Alps every year. Two hours after sitting down, I still hadn’t started the one thing I actually needed to do that day.
The frustrating part: I know every productivity framework in the book. Dorofeev, 4DX, GTD, Deep Work. I can diagram them on a whiteboard. None of them survives first contact with the reality of running operations across multiple countries and time zones.
So I built my own solution. Not by hiring engineers or buying another SaaS platform. I vibe-coded it myself using AI-assisted development. The result is a personal AI Chief of Staff with 77 integrated tools, 19 automated schedulers, a Telegram bot that turns voice messages into structured tasks, and a React dashboard that pulls everything into one view. I’m a MIPT graduate with a technical education, but the last time I wrote actual code was 15 years ago. This is the story of what I built, what broke, and what it means for any leader drowning in operational noise.
The real problem isn’t a lack of tools
An IBM study found that 84% of CMOs say fragmented operations limit their ability to use AI effectively. That number resonated because the bottleneck I kept hitting wasn’t strategic. It was operational. I was struggling to reach a point where I could make any decision at all because the information I needed was scattered across six different places.
The obvious answer is to buy something that integrates everything. But after 15 years in marketing at giant tech companies, I’ve learned that off-the-shelf tools are built for generic users. A leader managing seasonal demand spikes across four Alpine countries, coordinating drivers and partnerships and campaigns simultaneously, is not a generic user.
Building it, one failure at a time
The project started the way most AI experiments start these days: with ChatGPT. Useful for ten minutes, then the session ended, the context vanished, and I’d start over. Everyone who’s tried to use chat-based AI for real work has hit this wall.
The first breakthrough came with Cursor, an AI coding assistant with a persistent memory file. Every session, the AI would read a document containing my team structure, active projects, and priorities. Responses stopped being generic and became specific to my situation. But the system only worked when my laptop was open.
The second iteration changed everything. I built a Telegram bot and now have a voice message in, structured task out. Email draft in ten seconds. The first time I dictated a rambling three-minute voice note and watched it come back as a prioritised task list with a draft email, I knew this was fundamentally different. That bot grew to over 5,200 lines of code I never wrote in the traditional sense. I described what I wanted, tested the results, and guided the AI through revisions.
The third version connected all three interfaces into a single system. Telegram for mobile, Cursor for deep work, and an eight-page React dashboard for the full picture, including unified inbox, team status, projects, finance, and meetings. I use seventy-seven tools, nineteen schedulers that were built over weeks.
Why this isn’t just another chatbot
The most important design decision had nothing to do with technology. It was a rule embedded into the system’s core: interview before output. The AI never gives recommendations until it has asked me a structured set of questions first.
I loaded eleven decision-making frameworks into the system. Dorofeev for prioritisation. 4DX for focus. Premortem analysis for risk. Stakeholder simulation for pressure-testing decisions. Before the system produces any plan, three internal roles (architect, project manager, and domain stakeholder) critique it silently. Every output gets stress-tested in three rounds before it reaches me.
Most people build AI to do tasks. The real leverage comes from building AI that challenges your thinking before you act.
What broken
A voice transcription tool misrecognized a colleague’s name. The AI built an entire response around the wrong person. I caught it before it went out. So the lesson is that every automated output needs a human checkpoint.
During a routine update, the AI coding assistant overwrote a configuration file containing the system’s own API keys. Everything went dark. The environment file is now read-only.
The system once decided to parse my entire personal email inbox through an API. I found out when credit card notifications started flooding my phone. Monthly costs now sit at about 250 dollars, a fraction of a developer’s salary, but automated systems need spending guardrails as much as they need security ones.
Is it risky for someone who hasn’t coded in 15 years to build production systems? Fair question. The system runs behind an authentication proxy on a private cloud. Every code change goes through multi-role review. Nothing sends without my approval. It amplifies judgment. It doesn’t replace it.
What this means for business leaders
A Retool survey found that 48% of non-engineers already ship production software. That number will grow. The gap between “I can describe what I want” and “I have a working system” keeps shrinking.
A split is happening among senior leaders. On one side: those who evaluate and buy AI tools. On the other: those who build their own, tailored to their workflows. The leaders who build will move faster, because they won’t wait for a vendor to understand their specific problems. For mid-market companies without million-dollar engineering budgets, this changes the competitive equation entirely.
The detail that surprises people most is that I last wrote code 15 years ago. But that’s the point. AI has lowered the threshold enough for dormant technical instincts to become useful again. If you once understood how systems work, vibe coding lets you pick up where you left off. The muscle memory is still there. The AI fills in everything you’ve forgotten.
Since deploying the full system, I’ve reclaimed roughly 10 hours per week that used to vanish into email triage and context-switching. Time spent on actual strategic work has increased by about 20% percent.
I still sit down on Monday mornings with the same inbox, the same Slack, the same Asana. But now, by the time I open my laptop, the system has already triaged my email, flagged the three items that genuinely need me, drafted responses to the rest, and prepared a briefing on the week ahead. I still make every decision. The noise just got quieter.
—
About the Author
Denis Elkin, CMO at Alps2Alps
Denis Elkin, Chief Marketing Officer at Alps2Alps, a travel-tech mobility company providing airport-to-resort transfer. The company builds and runs tech-enabled transportation systems that manage cross-border demand, capacity, pricing, routing, and real-time coordination. With an engineering background and 15+ years in performance marketing, Denis builds AI-driven automation and measurement systems at the intersection of go-to-market execution and customer operations.
